Cross-domain traffic flow prediction method based on continuous learning

Abstract

The present invention discloses a cross-domain traffic flow prediction method based on continuous learning, which belongs to the field of traffic flow prediction. The method includes: incremental training of a cross-domain traffic flow prediction model, so that the inter-regional shared knowledge submodule extracts common fusion information between different tasks, obtains the cross-domain traffic flow prediction model of the current training round, and performs the prediction task, and updates the corresponding region-specific knowledge submodule according to the prediction result and the parameter knowledge of the model. In the continuous learning framework, while integrating multiple regional knowledge through the inter-regional shared knowledge module, the regional-specific knowledge submodule isolates the general knowledge and the specific knowledge between the regions, thereby ensuring the adaptability of the prediction of the target region and improving the ability of the transportation department to predict traffic flow; at the same time, based on the continuous learning framework, the transmission of traffic flow prediction knowledge between regions is realized, thereby solving the problem of difficulty in modeling due to the lack of traffic flow data.

Description

Technical Field

[0001] The present invention relates to the technical field of cross-domain traffic flow prediction, and in particular to a cross-domain traffic flow prediction method based on continuous learning. Background Art

[0002] Traffic flow monitoring and prediction aims to monitor and predict traffic flow by collecting, analyzing and utilizing traffic data, so as to guide traffic management, optimize traffic organization and improve traffic efficiency. Figure 1 As shown, the traffic flow monitoring and prediction method includes the following steps: data collection and cleaning; building a traffic flow prediction model; judging whether the model is valid, if not, optimizing the model; if so, applying it, i.e. predicting the traffic flow; judging whether the update node is reached, if not, using the current model; if so, updating the model, judging whether the model is valid, if not, optimizing the model; if so, conducting a new round of application to monitor and predict the traffic flow; continuously updating and optimizing the model, making decisions and deployments, such as Figure 2 As shown in the figure, different decisions are applied according to the size of traffic flow (such as controlling traffic lights). When the traffic flow is large, decision deployment A is executed, and when the traffic flow is small, decision deployment B is executed. In the above method, various types of models are built based on historical data, including traditional time series models, machine learning models, and deep learning models, which have become one of the main means of predicting traffic flow. These models can capture the dynamic characteristics of regional traffic and predict future traffic conditions, thereby providing decision support and reference for traffic management departments.

[0003] However, when the historical traffic data of a region is insufficient or incomplete, the traffic flow prediction model may face challenges in accuracy and stability. Existing traffic flow prediction schemes are almost all proposed for situations where there is sufficient data. How to effectively carry out traffic flow prediction when the amount of data is insufficient remains to be solved. At present, an important topic in the research of "smart transportation" is how to solve the challenges brought by insufficient data to traffic flow prediction, and through the innovation of technical means and methodology, to achieve accurate monitoring and reliable prediction of traffic flow, and provide more effective support for urban traffic management and planning. Summary of the invention

[0004] The purpose of the present invention is to overcome the problems of the prior art and provide a cross-domain traffic flow prediction method based on continuous learning.

[0005] The purpose of the present invention is to achieve the following technical solution: a cross-domain traffic flow prediction method based on continuous learning, the method comprising the following steps:

[0006] S1: constructing and initializing a cross-domain traffic flow prediction model; the cross-domain traffic flow prediction model includes a regional shared knowledge submodule and a regional exclusive knowledge submodule;

[0007] S2: Training the cross-domain traffic flow prediction model based on the training data set of the current training round, including the following sub-steps:

[0008] S21: Determine whether the training is the first training task, if so, execute step S22, otherwise, execute step S23;

[0009] S22: preprocessing the existing sample data to construct a training data set for the first training round; using the training data set for the first training round to train the initialized cross-domain traffic flow prediction model to obtain the cross-domain traffic flow prediction model corresponding to the first training round; the sample data is any one or more of image data, video data, GPS data, time series data, and public transportation data;

[0010] S23: preprocessing the newly added sample data to obtain a training data set for the current training round; using the training data set for the current training round, incrementally training the cross-domain traffic flow prediction model obtained in the previous training round, and in the incremental training process, using the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained in the previous training round to perform parameter optimization, so that the inter-regional shared knowledge submodule extracts common fusion information between traffic prediction tasks in different regions, and obtains the cross-domain traffic flow prediction model corresponding to the current training round;

[0011] S3: Use the current cross-domain traffic flow prediction model obtained through training to perform the cross-domain traffic flow prediction task, and then output the cross-domain traffic flow prediction result. According to the predicted traffic flow information of each region, combined with the parameter knowledge of the cross-domain traffic flow prediction model of the current training round, update the corresponding region-specific knowledge sub-module, so that the region-specific knowledge sub-module can extract the unique traffic flow feature information of each region.

[0012] In one example, the cross-domain traffic flow prediction model is any one of a fully connected neural network, a recurrent neural network, a graph neural network, a long short-term memory network, a Transformer model, a temporal graph convolution, and a multi-scale convolutional network.

[0013] In one example, parameter optimization is performed using parameter knowledge extracted from the cross-domain traffic flow prediction model obtained from the previous training round, including the following steps:

[0014] The parameters of the cross-domain traffic flow prediction model in the current round of training tasks are compared with the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained from the previous round of training tasks, and the parameter differences obtained from the comparison are incorporated into the loss function of parameter optimization in the current round of model training tasks, so that the parameters with larger weights in the cross-domain traffic flow prediction model obtained in the previous round of training tasks maintain larger weights in the loss function of the current round of training tasks.

[0015] In one example, after obtaining the cross-domain traffic flow prediction model corresponding to the first training round and / or the current training round, the method further includes:

[0016] Extracting parameter knowledge of cross-domain traffic flow prediction models.

[0017] In one example, an elastic weight consolidation method is used to extract parameter knowledge of a cross-domain traffic flow prediction model.

[0018] In one example, updating the corresponding region-specific knowledge submodule includes the following substeps:

[0019] If the current region is predicted, the original task is executed;

[0020] If a new region is predicted, a new region-specific knowledge sub-module is constructed based on the new parameters of the cross-domain traffic flow prediction model obtained from the current round of training while retaining the original task.

[0021] In one example, the process of training the cross-domain traffic flow prediction model also includes:

[0022] Determine whether the update cycle of the cross-domain traffic flow prediction model has been reached. If so, proceed to step S23 and enter the next round of training; otherwise, proceed to step S3.

[0023] In one example, when a new area appears or a fixed time period is reached, it is determined that the cross-domain traffic flow prediction model has reached an update period.

[0024] It should be further explained that the technical features corresponding to the above examples can be combined or replaced with each other to form a new technical solution.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. In one example, by setting up an inter-regional shared knowledge module and a region-specific knowledge sub-module in a continuous learning framework, the inter-regional shared knowledge module integrates the knowledge of multiple regions, while the region-specific knowledge sub-module isolates the common knowledge and the specific knowledge between the regions, thereby ensuring the adaptability of the prediction to the target area and improving the ability of the transportation department to predict traffic flow; at the same time, based on the continuous learning framework, the transfer of traffic flow prediction knowledge between regions is realized, thereby solving the problem of difficulty in modeling due to the lack of traffic flow data; further, through the incremental model update strategy, the present invention does not need to train the entire model from scratch, while ensuring the model performance, it avoids the resource consumption caused by re-modeling, thereby promoting "smart transportation" while submitting a "green transportation" plan.

[0027] 2. In one example, the elastic weight consolidation (EWC) method is used to extract the parameter knowledge of the model. By suppressing the change range of important parameters, the knowledge learned in the previous task is transferred to the next task. When the amount of data in the target area is insufficient, the cross-domain traffic flow knowledge is transmitted to provide an effective solution for predicting traffic flow in the target area, improve the generalization ability of the model, and enable the model to effectively adapt to traffic flow prediction tasks in new areas with limited sample data, thereby ensuring the accuracy and reliability of the prediction.

[0028] 34. In one example, when predicting a new area, the original task is retained, and a new region-specific knowledge sub-module is constructed based on the new parameters of the cross-domain traffic flow prediction model obtained from the current round of training. While updating the region-specific knowledge sub-module, catastrophic forgetting can be avoided, thereby adapting to the traffic flow prediction task in the new area and ensuring the model's prediction performance for the new area task. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The specific implementation methods of the present invention are further described in detail below in conjunction with the accompanying drawings. The accompanying drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The same reference numerals are used in these drawings to represent the same or similar parts. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0030] Figure 1 It is a flow chart of the existing traffic flow prediction method;

[0031] Figure 2 This is a framework diagram of the traffic flow prediction method;

[0032] Figure 3 A method flow chart provided for an example of the present invention;

[0033] Figure 4 A schematic diagram of a continuous learning dynamic expansion prediction model provided as an example of the present invention. DETAILED DESCRIPTION

[0034] The technical solution of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, "connected" and "connection" should be understood in a broad sense, for example, it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0036] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0037] In one example, if Figure 3 As shown, a cross-domain traffic flow prediction method based on continuous learning includes the following steps:

[0038] S1: Build and initialize a cross-domain traffic flow prediction model.

[0039] Among them, the cross-domain traffic flow prediction model is any one of a fully connected neural network, a recurrent neural network, a graph neural network, a long short-term memory network, a Transformer model, a temporal graph convolution, and a multi-scale convolutional network. The cross-domain traffic flow prediction model in this example is a fully connected neural network, including an input layer, several hidden layers, and an output layer. The output layer can be one or more fully connected layers. For the input object , using the bias vector and the weight matrix After a series of activation and linear processing, the final result is output by the output layer.

[0040] Furthermore, the cross-domain traffic flow prediction model includes a regional shared knowledge submodule and a regional specific knowledge submodule. The cross-domain traffic flow prediction model is trained based on sample data from multiple regions, and weights are shared between one or more fully connected layers, thereby extracting common fusion information between traffic prediction tasks in different regions, that is, the regional shared knowledge submodule is used to obtain common fusion information between traffic prediction tasks in different regions. The cross-domain traffic flow prediction model is trained based on sample data from a single region, and weights are not shared between one or more fully connected layers, thereby extracting unique traffic flow feature information for each region, that is, the regional specific knowledge submodule is used to obtain unique traffic flow feature information for each region.

[0041] Furthermore, the initialization of the cross-domain traffic flow prediction model includes weight and bias initialization, activation function selection, learning rate and optimizer selection, etc.

[0042] S2: The cross-domain traffic flow prediction model is trained based on the training data set of the current training round. The model training of the present invention adopts a continuous learning strategy. Each round of model training is a training task. The entire continuous learning training process includes the following sub-steps:

[0043] S21: Determine whether the training is the first training task, if so, execute step S22, otherwise, execute step S23.

[0044] S22: Preprocess the current sample data to construct a training data set for the first training round; use the training data set for the first training round to train the initialized cross-domain traffic flow prediction model to obtain the cross-domain traffic flow prediction model corresponding to the first training round.

[0045] The sample data is any one or more of image data, video data, GPS data, time series data, and public transportation data. Image data can be traffic image data of each road section including timestamp information; video data is road traffic video data including timestamp information; GPS data is GPS positioning data of taxis, buses, private vehicles, etc.; time series data includes numerical data such as historical traffic flow and speed recorded over time; public transportation data is the operation data of public transportation systems such as buses and subways, such as departure time and passenger capacity. The sample data is preprocessed, and the preprocessing includes processing missing data, deleting redundant features, constructing new features, flipping, cutting, scaling, etc. of image data, and the preprocessing method is used to prepare for training high-quality models. Furthermore, based on the training sample set, the back propagation method is used to train the cross-domain traffic flow prediction model to obtain the cross-domain traffic flow prediction model corresponding to the first task.

[0046] Preferably, after obtaining the cross-domain traffic flow prediction model corresponding to the first task, the EWC method is used to extract the parameter knowledge of the model. In the case of insufficient data, the model knowledge is represented by the importance of the parameters in the model through the EWC method, so that the information of the importance of the model parameters is transferred between models. As this knowledge is passed to the next task, the variation range of important parameters is limited, and this knowledge can be referenced when guiding the model training of the next task.

[0047] Preferably, it is determined whether the update cycle of the cross-domain traffic flow prediction model has been reached. If so, the process proceeds to step S23 to implement continuous training of the model, otherwise the process proceeds to step S3.

[0048] Optionally, the model update cycle can be adjusted according to the specific circumstances of the environment, and the time point of the update cycle can be preset, and the model update can be performed after the triggering time point. Specifically, when the model update cycle set by the target transportation department arrives, the newly added sample data that can be used for training during the update cycle time period is preprocessed to obtain the training data set for the current model training task. The update cycle here can be a fixed time period (such as 1 week or 1 month). However, when there is a situation where there is a large amount of new sample data in a certain cycle and a small amount of new sample data in another cycle, the difference in the number of sample data obtained in different fixed time periods may be large. In order to update the model in time to obtain more accurate predictions, an update cycle can be changed to the time when the current sample data that can be used for incremental training reaches a certain amount of data. Optionally, it can also be updated immediately when a new area appears, and the model can be continuously trained.

[0049] S23: Preprocess the newly added sample data to obtain the training data set of the current training round; use the training data set of the current training round to perform incremental training on the cross-domain traffic flow prediction model obtained in the previous training round, and during the incremental training process, use the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained in the previous training round to perform parameter optimization, so that the inter-regional shared knowledge sub-module extracts the common fusion information between traffic prediction tasks in different regions, and obtains the cross-domain traffic flow prediction model corresponding to the current training round.

[0050] Preferably, the parameter knowledge used in performing parameter optimization is extracted from the cross-domain traffic flow prediction model obtained from the previous training task, specifically:

[0051] Compare the parameters of the model in the current training task with the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained from the previous task, and incorporate the parameter differences obtained from the comparison into the loss function of the model parameter optimization of the current task, so that the parameters with larger weights in the cross-domain traffic flow prediction model obtained in the previous task have larger weights in the loss function of the current task. When minimizing the loss function of the current task, the difference between the parameters of the new and old tasks is taken into consideration in the loss function, limiting the range of parameter changes. The more important the parameters in the previous task, the greater the weight they have in the current loss function, and compared with general parameters, they have a greater impact on the loss function of the current task. By minimizing the loss function, the range of changes in different parameters is strictly limited, thereby retaining the knowledge learned in the previous task.

[0052] Taking two tasks A and B as examples, the training data sets of the two tasks are and After Task A builds a cross-domain traffic flow prediction model, the model knowledge is transferred to Task B through EWC. Weight information of model parameters As knowledge, it is extracted by task A and passed to task B through EWC, limiting the parameter weight information of task B The specific principle of constructing the loss function is introduced as follows:

[0053] ① According to the Bayesian formula, by using the parameters The prior probability , marginal probability , conditional probability , the posterior probability can be calculated by the following formula :

[0054] ;

[0055] ② Since the dataset D is completely divided into task A and task B, the above formula is transformed into formula (3):

[0056] ;

[0057] in, Indicates the task The conditional probability of Indicates the task The posterior probability of Indicates the task The marginal probability of .

[0058] ③Since each subset affects the posterior probability, according to the Laplace approximation method, the loss function of EWC can be expressed as:

[0059] ;

[0060] in, represents the total loss; It only represents the loss of Task B; Based on the data set The Laplace approximation of the posterior probability is calculated; is the relative importance of the task; Indicates the current model parameters. The value of the parameter; Indicates that in the task The parameters learned The value of .

[0061] ④ When Task B builds a model, the loss of parameters in Task B is part of the total loss, and the parameters are relative to the parameters of Task A. The change range of is the other part of the total loss. The final selected parameter is to make The parameters with the minimum value, ideally, perform better on Task B; at the same time, the more important parameters in Task A have a smaller change compared to the current parameters.

[0062] Preferably, after obtaining the cross-domain traffic flow prediction model corresponding to the current training round, the EWC method is used to extract the parameter knowledge of the model so that the information of the importance of the model parameters is transferred between models to guide the model training for the next task.

[0063] S3: Using the current cross-domain traffic flow prediction model obtained through training, perform the cross-domain traffic flow prediction task, including the following sub-steps:

[0064] S31: For target transportation departments, in areas with insufficient traffic flow data, input existing sample data, such as road traffic images and / or time series data (numerical data such as historical traffic flow and speed recorded over time) of each road section and time point in the target area, and use the cross-domain traffic flow prediction model of the current training round obtained through training to output the cross-domain traffic flow prediction result;

[0065] S32: Based on the traffic flow information of each region in the traffic forecast, combined with the preset parameter knowledge of the cross-domain traffic flow forecast model obtained in the current training round, the region-specific knowledge sub-module corresponding to the predicted traffic flow information is updated, so that the region-specific knowledge sub-module extracts the unique traffic flow feature information of each region, including the following sub-steps:

[0066] If the current region is predicted, the original task is executed;

[0067] If a new region is predicted, a new region-specific knowledge submodule is constructed based on the new parameters of the cross-domain traffic flow prediction model obtained from the current round of training while retaining the original task. Under the continuous traffic flow prediction architecture, as the number of regions increases, the region-specific knowledge submodule increases dynamically, and the inter-regional shared knowledge submodule can integrate more common knowledge from different regions. This example constructs subnetworks based on parameters of different regions, and different tasks can use the same main module and call different submodules, so as to update the submodules and avoid catastrophic forgetting.

[0068] In the process of training the cross-domain traffic flow prediction model, the present invention first constructs and initializes the cross-domain traffic flow prediction model, and then takes each round of training model as a training task, and trains the model by constructing a training data set for the current training round, wherein, for the first training task, the current existing traffic flow sample data is directly preprocessed to obtain the corresponding training data set, and for subsequent other training tasks, the corresponding training data set is obtained by preprocessing the newly added sample data within the update cycle (the time interval between the previous training round and the current training round). And, after each round of training is completed, the extracted model parameter knowledge is passed to the next task. In the process of training the model, the next task combines the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained from the previous training task to perform parameter optimization, so that the model is continuously updated, thereby completing the continuous learning process of the model.

[0069] Specifically, the present invention adopts a continuous learning strategy to train a cross-domain traffic flow prediction model. As the number of regions increases, the existing common knowledge is extracted and transmitted to the next stage, thereby incrementally updating the inter-regional shared knowledge modules between regions. In addition, considering the differences between the reference region and the target region, the present invention allocates an exclusive sub-module structure to each region to learn and store the exclusive knowledge of each region. As the number of reference regions increases, region-specific sub-modules are added accordingly to preserve region-specific prediction knowledge. At the same time, each reference region performs knowledge fusion to continuously enhance the prediction performance of the model. Therefore, the method of incrementally adding exclusive modules and continuously updating shared modules and inter-regional shared knowledge modules proposed in the present invention not only avoids knowledge interference between regions, but also helps regions with insufficient traffic flow samples to effectively carry out traffic flow prediction tasks by sharing common knowledge. Figure 4As shown, the present invention adopts an incremental modeling approach to build a model with the help of existing knowledge. In order to predict the traffic flow of the target area (X) with insufficient sample data, the model of multiple areas (A, B, C, etc.) with sufficient data is used to continuously integrate the model knowledge of multiple areas through the inter-regional shared knowledge sub-module, helping the target area to overcome the modeling problem caused by insufficient data, making the prediction model performance more and more stable and the prediction effect more and more accurate. At the same time, based on the inter-regional shared knowledge module, the regional specific knowledge sub-module corresponding to the target area is trained in combination with the local available data, which not only improves the prediction performance of the model by borrowing general knowledge, but also retains local specific information to enhance the applicability of the model in the target area.

[0070] The specific implementation methods are detailed descriptions of the present invention. It cannot be considered that the specific implementation methods of the present invention are limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions and substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A cross-domain traffic flow prediction method based on continuous learning, characterized in that: The following steps are involved: S1: constructing and initializing a cross-domain traffic flow prediction model; the cross-domain traffic flow prediction model includes an inter-regional shared knowledge submodule and a region-specific knowledge submodule, the inter-regional shared knowledge submodule is used to obtain common fusion information between traffic prediction tasks in different regions, and the region-specific knowledge submodule is used to obtain unique traffic flow feature information for each region; S2: Training the cross-domain traffic flow prediction model based on the training data set of the current training round, including the following sub-steps: S21: Determine whether the training is the first training task, if so, execute step S22, otherwise, execute step S23; S22: Preprocess the current sample data to construct the training data set for the first training round; Using the training data set of the first training round, the initialized cross-domain traffic flow prediction model is trained to obtain the cross-domain traffic flow prediction model corresponding to the first training round; the sample data is any one or more of image data, video data, GPS data, time series data, and public transportation data; S23: preprocess the newly added sample data to obtain the training data set of the current training round; Using the training data set of the current training round, the cross-domain traffic flow prediction model obtained in the previous training round is incrementally trained. In the incremental training process, the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained in the previous training round is used to perform parameter optimization, so that the inter-regional shared knowledge submodule extracts the common fusion information between traffic prediction tasks in different regions, and obtains the cross-domain traffic flow prediction model corresponding to the current training round; S3: Using the current cross-domain traffic flow prediction model obtained through training, perform the cross-domain traffic flow prediction task, and then output the cross-domain traffic flow prediction result. According to the predicted traffic flow information of each region, combined with the parameter knowledge of the cross-domain traffic flow prediction model of the current training round, update the corresponding region-specific knowledge sub-module, so that the region-specific knowledge sub-module extracts the unique traffic flow feature information of each region; Update the corresponding region-specific knowledge submodule, including the following substeps: If the current region is predicted, the original task is executed; If a new region is predicted, a new region-specific knowledge submodule is constructed based on the new parameters of the cross-domain traffic flow prediction model obtained from the current round of training while retaining the original task; The training process of the cross-domain traffic flow prediction model also includes: Determine whether the update cycle of the cross-domain traffic flow prediction model has been reached. If so, proceed to step S23 and enter the next round of training; otherwise, proceed to step S3; When a new area appears, it is determined that the cross-domain traffic flow prediction model has reached the update cycle.

2. The cross-domain traffic flow prediction method based on continuous learning according to claim 1 is characterized in that: The cross-domain traffic flow prediction model is any one of a fully connected neural network, a recurrent neural network, a graph neural network, a long short-term memory network, a Transformer model, a temporal graph convolution, and a multi-scale convolutional network.

3. The cross-domain traffic flow prediction method based on continuous learning according to claim 1 is characterized in that: Parameter optimization is performed using the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained from the previous training round, including the following steps: The parameters of the cross-domain traffic flow prediction model in the current round of training tasks are compared with the parameter knowledge extracted from the cross-domain traffic flow prediction model obtained from the previous round of training tasks, and the parameter differences obtained from the comparison are incorporated into the loss function of parameter optimization in the current round of model training tasks, so that the parameters with larger weights in the cross-domain traffic flow prediction model obtained in the previous round of training tasks maintain larger weights in the loss function of the current round of training tasks.

4. The cross-domain traffic flow prediction method based on continuous learning according to claim 1 is characterized in that: After obtaining the cross-domain traffic flow prediction model corresponding to the first training round and / or obtaining the cross-domain traffic flow prediction model corresponding to the current training round, the method further includes: Extracting parameter knowledge of cross-domain traffic flow prediction models.

5. The cross-domain traffic flow prediction method based on continuous learning according to claim 4 is characterized in that: The elastic weight consolidation method is used to extract parameter knowledge of the cross-domain traffic flow prediction model.

6. The cross-domain traffic flow prediction method based on continuous learning according to claim 1 is characterized in that: When a new region appears, the cross-domain traffic flow prediction model is judged to have reached the update cycle, which is replaced by: When a fixed time period is reached, it is determined that the cross-domain traffic flow prediction model has reached the update period.

Images